Learning the underlying patterns in the data goes beyond instance-based generalization to some external knowledge represented in structured graphs or networks. Deep Learning (DL) has shown significant advances in probabilistically learning latent patterns in the data using a multi-layered network of computational nodes (i.e. neurons/hidden units). However, with the tremendous amount of training data, uncertainty in generalization on domain-specific tasks, and delta improvement with an increase in complexity of models seem to raise a concern on the features learned by the model. As incorporation of domain specific knowledge will aid in supervising the learning of features for the model, infusion of knowledge from knowledge graphs within hidden layers will further enhance the learning process. Although much work remains, we believe that KGs will play an increasing role in developing hybrid neuro-symbolic intelligent systems (that is bottom up deep learning with top down symbolic computing) as well as in building explainable AI systems for which KGs will provide a scaffolding for punctuating neural computing. In this position paper, we describe our motivation for such hybrid approach and a framework that combines knowledge graph and neural networks.

This article reviews recent advances in multi-agent reinforcement learning algorithms for large-scale control systems and communication networks, which learn to communicate and cooperate. We provide an overview of this emerging field, with an emphasis on the decentralized setting under different coordination protocols. We highlight the evolution of reinforcement learning algorithms from single-agent to multi-agent systems, from a distributed optimization perspective, and conclude with future directions and challenges, in the hope to catalyze the growing synergy among distributed optimization, signal processing, and reinforcement learning communities.

In this paper, we present a technique that improves the process of training an agent (using RL) for instruction following. We develop a training curriculum that uses a nominal number of expert demonstrations and trains the agent in a manner that draws parallels from one of the ways in which humans learn to perform complex tasks, i.e by starting from the goal and working backwards. We test our method on the BabyAI platform and show an improvement in sample efficiency for some of its tasks compared to a PPO (proximal policy optimization) baseline.

Commencing in the 1980s with the appearance of specialised isolated-word recognition (IWR) systems for military command-and-control equipment, spoken language technology has evolved from large-vocabulary continuous speech recognition (L VCSR) for dictating documents (such as Dragon's Naturally Speaking and IBM's Via V oice) released in the late 1990s, through telephone-based interactive voice response (IVR) systems to the launch of Siri (Apple's voice-enabled personal assistant for the iPhone) in 2011. Siri was quickly followed by Google Now and Microsoft's Cortana. The following years heralded a new era of smart speaker based voice assistants, starting with Amazon's 2015 release of Alexa followed later by Google Home, Apple's HomePod and Sonos One. These contemporary systems not only represent the successful culmination of over 50 years of laboratory-based speech technology research (Pieraccini, 2012), but also signify that speech technology had finally become "mainstream" (Huang, 2002) (at least, in the English-speaking world). Indeed, the market penetration of these smartphone and smart speaker based voice assistants is astounding. For example, Siri has had over 40 million monthly active users in the U.S. since July 2017, Google Assistant is available on over 225 home-control brands and more than 1,500 devices, and tens of millions of Alexa-enabled devices were sold worldwide over the 2017 Christmas holiday season (Boyd, 2018). Also, a study by Juniper Research (Smith, 2017) estimated that the number of voice assistant devices across all Figure 1: The evolution of spoken language processing applications from specialised military'command-and- control' systems of the 1980/90s to contemporary'voice-enabled personal assistants' (such as Siri and Alexa) and future'autonomous social agents', i.e. robots.

Tatiana Mejia, Head of AI Product Marketing and Strategy at Adobe and named one of 2018's top Silicon Valley women in AI, joins us to talk about the present and future of machine learning tools for digital designers.

Also included is an essay from SIAM News'The Functions of Deep Learning' (December 2018) A second distributor for SIAM members is siam.org We will confirm orders for this new book by email.

As part of Stanford's ongoing 100-year study on artificial intelligence, known as the AI100, two workshops recently considered the issues of care technologies and predictive modeling to inform the future development of AI technologies. "We are now seeing a particular emphasis on the humanities and how they interact with AI," said Russ Altman, Stanford professor of engineering and the faculty director of the AI100. The AI100 is project of the Stanford Institute for Human-Centered Artificial Intelligence. After the first meeting of the AI100, the group planned to reconvene every five years to discuss the status of the AI industry. The idea was that reports from those meetings would capture the excitement and concerns regarding AI technologies at that time, make predictions for the next century and serve as a resource for policymakers and industry stakeholders shaping the future of AI in society.

Long-time Slashdot reader clawsoon writes: Among Sunday's releases from the International Consortium of Investigative Journalists on leaked Chinese documents about the detention of Xinjiang Uighurs -- which they are calling the largest mass internment of an ethnic-religious minority since World War II -- is a section on detention by algorithm which "is more than a'pre-crime' platform, but a'machine-learning, artificial intelligence (AI), command and control' platform that substitutes artificial intelligence for human judgment...." "The Chinese have bought into a model of policing where they believe that through the collection of large-scale data run through AI and machine learning that they can, in fact, predict ahead of time where possible incidents might take place, as well as identify possible populations that have the propensity to engage in anti-state anti-regime action," reports James Mulvenon, director of intelligence integration at SOS International LLC, an intelligence and information technology contractor for several U.S. government agencies. "And then they are preemptively going after those people using that data." The Chinese government responded by calling the leaked documents "fake news."

Video-to-video synthesis (vid2vid) aims at converting an input semantic video, such as videos of human poses or segmentation masks, to an output photorealistic video. While the state-of-the-art of vid2vid has advanced significantly, existing approaches share two major limitations. Numerous images of a target human subject or a scene are required for training. Second, a learned model has limited generalization capability. A pose-to-human vid2vid model can only synthesize poses of the single person in the training set.

Success in any area is often a combination of three things: talent, hard work and perseverance. For software-as-a-service (SaaS) company Darts-ip, all three were needed to grow a pioneering idea from a handful of people to a 300-strong organisation in just 13 years. The talent came in the form of two groups from very different industries. The service they wanted to offer, to make legal research as easy as possible, came from trademark lawyer and Darts-ip founder Jean-Jo Evrard. While working in Brussels and Paris for law firm NautaDutilh, Evrard was frustrated.